At SC19, the Association for Computing Machinery (ACM) awarded the prestigious Gordon Bell Prize to the Swiss Federal Institute of Technology (ETH) Zurich. The team’s project, “A Data-Centric Approach to Extreme-Scale Ab Initio Dissipative Quantum Transport Simulations,” had leveraged a new programming framework and the most powerful publicly ranked supercomputer to shine a light on the intricate behaviors of transistors.
Modern transistors are typically around a thousand times smaller than extremely fine sand, allowing billions of them to fit on computer chips. Transistors also heat up as they handle electric signals, and as transistors become denser, so does the heat that needs to be dissipated.
“If we want to mitigate this problem, we need to understand where this heat comes from, how it’s generated, and how we can better evacuate it from the active regions of these transistors,” said Mathieu Luisier, one of the researchers and an associate professor at ETH Zurich. “Simulations can guide us toward the knowledge of where heat is generated and specifically how it’s dependent on the current that flows.”
A full 3D transistor contains about a million atoms – far beyond what modern supercomputers could effectively simulate. So the team, led by Torsten Hoefler (another associate professor at ETH Zurich), used Summit, the world’s fastest publicly ranked supercomputer, to conduct a simulation of a small, 2D cross-section of a transistor with a resolution of 10,000 atoms. “We wanted this to be realistic and reach the dimensions of manufactured transistors,” Luisier said. “This corresponds to what is actually fabricated.”
To run such an intensive simulation, they used a new data-centric (or “DaCe”) iteration of the OMEN nanodevice simulator. This DaCe framework, developed at ETH Zurich, allowed programmers to optimize the simulation more dynamically, rather than going line-by-line through the code. DaCe sped the 10,000-atom simulation up to just eight minutes – 14 times faster than previous 1,000-atom simulations. Of course, Summit helped, too. “Even just the memory requirements of our program require a resource like Summit,” Hoefler said.
The team hopes that the DaCe framework will benefit semiconductor production. “We want to be ready to model a design if a company comes to us and wants to try something new,” Luisier said. “If we can directly construct a simulation of a design, we could tell them how much the temperature will increase based on this design, and that could help them create better transistors.” They also anticipate applications far beyond nanoelectronics.
For their revolutionary work, the team from ETH Zurich won 2019’s Gordon Bell Prize, which recognizes outstanding achievements in high-performance computing. The prize, which comes with a $10,000 award, was announced at SC19 yesterday in Denver, Colorado, and presented to the team by Cherri Pancake, ACM President, and Arndt Bode, chair of the 2019 Gordon Bell Prize Award Committee. The other finalists were a team from the University of Michigan that also used Summit – in that case, to perform a large, highly accurate alloys simulation.